인공 신경망 기법을 이용한 크리프-피로 수명예측 방법에 관한 연구 = (A) study of the creep-fatigue life prediction method using artificial neyral network

다국어 초록 (Multilingual Abstract)

Stainless steels have good high temperature mechanical properties and are generally used for high temperature components such as pressure vessels, boilers and reactors in nuclear power plants. Although it differs to some degree, most of time, these components are exposed to a loading condition where the static and the dynamic loads are applied repeatedly or simultaneously. Such conditions at high temperature may cause early fracture due to crack generation, propagation and material damage by fatigue and creep deformation.
Also, major high temperature components in steam power plants such as turbines and boilers can be severely damaged due to creep-fatigue interaction, which is composed of creep damage by the static load and fatigue damage by the thermal stress generated due to temperature changes during starting and stopping.
In the early analytical research works on the creep-fatigue interaction, the parameters used for the characterization of the pure fatigue crack propagation were usually employed. However, some of the factors were found to be improper for the evaluation of the time dependent behavior in the static loading regime. Therefore, in later analysis works, the linear cumulative damage rule was usually employed which considers creep damage and fatigue damage separately, excluding the effect of creep-fatigue interaction.
Although many studies on this subject, including the effect of vacancy, crack tip blunting, and oxidation, have been reported, the explanation of creep-fatigue interaction is not clear yet. Due to this reason, it is still difficult to evaluate the remaining life of high temperature components. Until now, both theoretical and experimental equations have been proposed for the life prediction and when the results of these equations do not agree well with the experimental results, the proposed equations were modified. However, even the modified equations could give only limited predictions. Artificial neural networks are computational models that are designed to mimic human brain architecture and operations. The networks have shown remarkable performance when used to model complex linear and nonlinear relationships, especially in the fields of signal processing, non-destructive testing, corrosion life prediction and some other fields of materials science.
In this study, using AISI 316 stainless steel, creep-fatigue tests were carried out under various test conditions(different total strain ranges and hold times) to verify the applicability of the artificial neural network method to the creep-fatigue life prediction. Life prediction was made also by the Coffin-Manson, the modified Coffin-Manson method and the Ostergren, the modified Ostergren method besides the artificial neural network method using verification data points out of total experimental data points. The verification data points were carefully chosen for the purpose of evaluating the predictability of each method. The predicted lives were compared with the experimental results and the following conclusions were obtained within the scope of this study.
After conducting creep-fatigue life prediction by the Coffin-Manson, the modified Coffin-Manson method, the Ostergren, the modified Ostergren method and the artificial neural network method, the results were compared and the following conclusions were obtained within the scope of this study.
1. As a creep-fatigue life prediction method for the simultaneously applied tensile and compressive hold time condition, the artificial neural network method with adaptive learning rate proved to be far more effective and accurate than the Coffin-Manson, the modified Coffin-Manson method or the Ostergren, the modified Ostergren method. This superiority of the artificial neural network method comes from its ability of distinguishing the effect of tensile hold time from that of compressive hold time.
2. Compared with the pure fatigue life, the creep-fatigue life decreased significantly under tensile hold time. This tendency became more significant as total strain rahge and tensile hold time increases, which is the result of cavity generation at grain boundaries. Compared with the pure fatigue condition, as compressive hold time increased the creep-fatigue life was found to be decreasing under all tested conditions but the decrease was most significant under tensile hold time. However, slight decrease in the creep-fatigue life was observed under tensile+compressive hold time compared with the case under tensile hold time only. This is because the growth of cavity, generated under tensile hold time, is either interrupted or terminated during compressive hold time.
3. In all the tensile hold time tests, rapid stress relaxation occurs in first few seconds of the strain hold, followed by a slow rate of stress relaxation during the rest of the hold period. With the increase in duration of hold time, a amount of stress relaxation increased whereas the peak tensile stress reduced.

목차 (Table of Contents)

• 목차
• * Nomenclature = iv
• * List of Figures = vi
• * List of Tables = ix
• 제1장 서론 = 1
• 1.1 연구 배경 = 1
• 1.2 연구 목적 = 3
• 1.3 논문의 구성 = 3
• 제2장 연구현황 및 관련이론 = 5
• 2.1 연구 현황 = 5
• 2.1.1 크리프-피로 상호작용에 관한 연구 = 5
• 2.1.2 크리프-피로 손상기구에 관한 연구 = 7
• 2.1.3 크리프-피로 상호작웅시 파단양상에 관한 연구 = 8
• 2.1.4 고온 저주기 피로 특성에 미치는 여러 변수의 영향 = 14
• 2.1.5 고온 저주기 피로실험시 형성된 전위구조에 관한 연구 = 18
• 2.1.6 크리프-피로 상호작용시의 수명예측 방법에 관한 연구 = 19
• 2.2 기존의 수명예측과 관련된 이론 = 24
• 2.2.1 Coffin-Manson에 의한 수명예측 접근법 = 24
• 2.2.2 Ostergren에 의한 수명 예측 접근법 = 26
• 2.3 개선된 인공신경망 기법에 의한 수명 예측법 = 27
• 2.3.1 인공 신경망의 개념 = 28
• 2.3.2 다층인식자 = 29
• 2.3.3 역전파 학습 규칙 = 31
• 제3장 실험 방법 및 결과 = 40
• 3.1 시편 재료 = 40
• 3.2 시편 열처리 = 40
• 3.3 인장실험 = 40
• 3.3.1 인장실험용 시편 = 40
• 3.3.2 인장 실험 조건 = 41
• 3.3.3 인장 실험 결과 = 41
• 3.4 크리프-피로 실험 = 41
• 3.4.1 크리프-피로 실험용 시편 = 41
• 3.4.2 크리프-피로 실험 조건 = 42
• 3.4.3 크리프-피로 실험 절차 = 42
• 3.4.4 크리프-피로 실험 결과 = 44
• 3.5 파단면 관찰 = 53
• 3.5.1 파단면 관찰 결과 = 54
• 제4장 기존의 예측방법에 의한 수명예측 = 56
• 4.1 인장 유지시간이 작용했을때의 수명예측 = 56
• 4.1.1 Coffin-Manson 접근법에 의한 수명예측 = 57
• 4.1.2 Modified Coffin-Manson 접근법에 의한 수명예측 = 59
• 4.1.3 Ostergren 접근법에 의한 수명예측 = 61
• 4.1.4 Modified Ostergren접근법에 의한 수명예측 = 63
• 4.1.5 인장유지시의 수명예측 결과 비교 = 65
• 4.2 인장-압축 유지시간이 작용했을때의 수명예측 = 66
• 4.2.1 Coffin-Manson 접근법에 의한 수명예측 = 66
• 4.2.2 Modified Coffin-Manson 접근법에 의한 수명예측 = 68
• 4.2.3 Ostergren 접근법에 의한 수명예측 = 70
• 4.2.4 Modified Ostergren접근법에 의한 수명예측 = 72
• 4.2.5 인장-압축유지시의 수명예측 결과 비교 = 75
• 4.3 인장+압축 유지시간이 작용했을때의 수명예측 = 76
• 4.3.1 Coffin-Manson 접근법에 의한 수명예측 = 76
• 4.3.2 Modified Coffin-Manson 접근법에 의한 수명예측 = 78
• 4.3.3 Ostergren 접근법에 의한 수명예측 =81
• 4.3.4 Modified Ostergren 접근법에 의한 수명예측 =82
• 4.3.5 인장·압축유지시의 수명예측 결과 비교 = 85
• 제5장 인공신경망 기법에 의한 수명예측 = 86
• 5.1 인장 유지시간이 작용했을때의 수명예측 = 86
• 5.1.1 적응 학습율을 적용한 인공신경망 기법에 의한 수명예측 = 86
• 5.1.2 적응 학습율을 적용하지 않은 인공신경망 기법에 의한 수명예측 = 93
• 5.1.3 인장유지시의 수명예측 결과 비교 = 96
• 5.2 인장-압축 유지시간이 작용했을때 의 수명예측 = 97
• 5.2.1 적응 학습율을 적용한 인공신경망 기법에 의한 수명예측 = 97
• 5.2.2 적웅 학습율을 적용하지 않은 인공신경망 기법에 의한 수명예측 = 104
• 5.2.3 인장-압축유지시의 수명예측 결과 비교 = 107
• 5.3 인장+압축 유지시간이 작용했을때 의 수명예측 = 108
• 5.3.1 적응 학습율을 적용한 인공신경망 기법에 의한 수명예측 = 108
• 5.3.2 적응 학습율을 적용하지 않은 인공신경망 기법에 의한 수명예측 = 115
• 5.3.3 인장+압축유지시의 수명예측 결과 비교 = 118
• 제6장 결론 및 향후 연구과제 = 119
• 6.1 결론 = 119
• 6.2 향후 연구과제 = 120
• 참고문헌 = 122
• ABSTRACT = 128